puzzle picture
2106: Design with Neural Net Objective
Status: Closed


Name: 2106: Design with Neural Net Objective
Status: Closed
Created: 02/09/2022
Points: 100
Expired: 02/16/2022 - 23:00
Difficulty: Intermediate
Description: Design a protein with the help of the new Neural Net Objective! The Neural Net Objective can help identify parts of your protein that should be mutated to improve AlphaFold prediction confidence. After you use the AlphaFold tool to get a prediction, click "Show Neural Net Objective" to color your protein based on local distogram analysis. To improve AlphaFold prediction confidence, try mutating residues that are colored red by the Neural Net Objective. Players should aim for an AlphaFold confidence and similarity of 80% or more.

We are especially interested to see what creative new folds players might come up with. There are no additional Objectives, but we expect successful designs will still need lots of helices or sheets, with short minimal loops, as well as a closely-packed core of orange hydrophobic residues. AlphaFold predictions will not affect your score. Players may insert and delete residues to a maximum of 120 residues total.
Categories: Design, Overall

Top Groups

2Anthropic Dreams13,43760
3Beta Folders13,26733
4Go Science13,22917
5Void Crushers13,1358

Top Evolvers

Top Soloists

Need this puzzle? Log in to download.  


bkoep's picture
User is online Online
Joined: 11/15/2012
Groups: Foldit Staff

Neural Net (max +0)
Achieve an AlphaFold prediction confidence of 80% or more.

Joined: 12/06/2008
Groups: Contenders
not understanding the Neural Net tool

My first attempt at a design gave a 52.6% confidence. When I clicked the "show neural net" option in the objectives, all 120 residues turn an identical shade of dark grey. Not one bit of red.

My design is awful. Why is the game telling me there is nothing to change?

bkoep's picture
User is online Online
Joined: 11/15/2012
Groups: Foldit Staff
Completely disordered?

That is a little odd if everything is exactly the same shade, and may indicate a bug. Can you share this solution with scientists?

But there could be other cases where the AlphaFold confidence is very low, and the Neural Net Objective is completely lukewarm.

Many sequences are not expected to fold at all, and in the real world these will adopt an "intrinsically disordered" or "random coil" structure. In such cases, we would expect AlphaFold to predict a relatively featureless distogram, where every residue-residue distance is represented as a flat distribution (i.e. "the distance between these two residues could be anything"). Probably the AlphaFold predicted structure will be very loopy and loosely-packed.

In this case, the distogram cross entropy will not be good, but also it may not be especially bad. We might interpret this like AlphaFold saying, "I don't see any evidence that your design will fold into this structure, but also I don't see any evidence that it will fold into anything else."

Joined: 12/06/2008
Groups: Contenders
User error?

Yes, everything is exactly the same dark shade of grey. I'll share, next time I log into the game.

I think I may be experiencing a "1D10T" error. I really don't understand how to use this tool.

Okay, I upload my solution to AlphaFold and wait for it to spit out some percentages. What am I supposed to do, next? Can y'all please give us some training?

Vincera's picture
User offline. Last seen 15 weeks 2 days ago. Offline
Joined: 02/14/2010
Groups: None
AF Db down?

For the past two hours or so, I've attempted (unsuccessfully) to upload to AF. My [circa] midnight uploads transmitted and rendered results but since this morning, nada. Are any other players experiencing this problem?

Joined: 03/28/2020
Groups: Go Science
yes, same thing here

AF worked perfectly in the past but since today I have the same problem. I guess the server is down/crashed.

bkoep's picture
User is online Online
Joined: 11/15/2012
Groups: Foldit Staff
AlphaFold server down

Yes, it looks like the AlphaFold server is currently down for everyone. We're looking into the issue.

Joined: 03/28/2020
Groups: Go Science
Very limited upload to scientists?

I have noticed in the last weeks that the amount of uploadable solutions to scientists is very limited. For this puzzle I can only upload 5 solutions. Is this intended? I would have more to share but I do get some message like "too many shared solutions..." I can upload many more to teammembers though. Is this a bug or a feature? I think this cropped up since AF was introduced but I am mot sure there.

bkoep's picture
User is online Online
Joined: 11/15/2012
Groups: Foldit Staff
Limited to 5 pending uploads

Yes, we currently limit AlphaFold submissions to maximum 5 pending requests at any time. Once a pending job completes, then you may submit a new job.

Since our AlphaFold server is currently down, pending jobs are not completing. Once we get the server back online and your pending jobs are completed, then you will be able to make new submissions. Apologies for the unexpected downtime!

Joined: 03/28/2020
Groups: Go Science
Sorry for being imprecise

I did not mean uploads to alphafold but instead the upload of a solution to scientists in the filedialog (where I could also upload to group or to myself). THIS is limited to 5 for me in shares to scientists. Or are the number of pending uploads to AF and the number of shareable solutions in filedialog coupled? I would find that strange.
Anyway, thx for trying to get the AF server back online. I will have to stall my investigations until it is available. Good luck in getting it back up again!

bkoep's picture
User is online Online
Joined: 11/15/2012
Groups: Foldit Staff
Share with scientist limits

Sorry, I misread your comment! Yes, players are limited to 5 shares with scientists in each puzzle; we will increase it to 10.

The purpose of the limit is to reduce spam, and discourage players from sharing with scientists every time they make a minor change to their solution.

The Share with Scientists limit is unrelated to the AlphaFold submission limit.

Joined: 03/28/2020
Groups: Go Science
Makes sense

It would be great to be able to share a bit more to scientists - especially for these kind of puzzles which are designed to try a multitude of different approaches. Thx!

Joined: 09/24/2012
Groups: Go Science
how to

I noticed that the higher my foldit points, the higher the AF confidence.

It's disappointing because this doesn't help me to select early designs with higher confidence. On end play, the AF doesn't help to select and finish with high confidence solutions (because most of them are high, and if a high score isn't, I still shall end with this for pure competition).

I think that a small bonus related to confidence (once you stop mutating) could help to align high score with confidence.

User login
Download links:
  Windows    OSX    Linux  
(10.12 or later)

Are you new to Foldit? Click here.

Are you a student? Click here.

Are you an educator? Click here.
Social Media

Only search fold.it
Other Games: Mozak
Recommend Foldit
Top New Users

Developed by: UW Center for Game Science, UW Institute for Protein Design, Northeastern University, Vanderbilt University Meiler Lab, UC Davis
Supported by: DARPA, NSF, NIH, HHMI, Amazon, Microsoft, Adobe, Boehringer Ingelheim, RosettaCommons